Estimating Employer Labour-Market Power and Its Wage Effects

Making their own weather? Estimating employer labour-market power and its wage effects∗ Pedro S. Martinsy Queen Mary University of London & NovaSBE & IZA & GLO December 30, 2019 Abstract The subdued wage growth observed over the last years in many countries has spurred renewed interest in monopsony views of the labour market. This paper is one of the first to measure the extent and robustness of employer labour-market power and its wage implications exploiting comprehensive matched employer-employee data. We find average (employment-weighted) Herfindhal indices of 800 to 1,100; and that less than 9% of workers are exposed to concentration levels thought to raise market power concerns. However, these figures can increase significantly with different methodological choices. Finally, when controlling for both worker and firm heterogeneity and instrumenting for concentration, wages are found to be negatively affected by employer concentration, with elasticities of around -1.5%. Keywords: Oligopsony, Wages, Portugal. JEL Codes: J42, J31, J63. ∗The author thanks Andrea Bassanini, Wenjing Duan, Judite Goncalves, Juan Jimeno, Claudio Lucifora, Ioana Marinescu, Bledi Taska and workshop participants at Queen Mary University of London and the Euro- pean Commission for comments. The author also thanks INE and the Ministries of Education and Employment, Portugal, for data access, and the European Union for financial support (grant VS/2016/0340). yEmail: [email protected]. Web: https://sites.google.com/site/pmrsmartins/. Address: School of Business and Management, Queen Mary University of London, Mile End Road, London E1 4NS, United Kingdom. Phone: +44/0 2078822708. 1 1 Introduction The limited wage growth observed in many countries in the recovery following the 2008 financial crisis has prompted important questions about the degree of wage setting power enjoyed by employers. These questions have gathered additional momentum with the complementary evidence that has emerged over the last decade on declining labour shares, the rise of 'su- perstar firms’ (Autor et al. 2017), the relevance of 'no poaching agreements' in employment contracts (Krueger & Ashenfelter 2018), and the limited disemployment effects of minimum wages. While these findings may still be reconciled with largely competitive labour markets, the relevance of evidence on employer labour-market power is clear. This paper contributes to the literature on monopsony (Staiger et al. 2010, Manning 2010, 2011, Falch 2010, Matsudaira 2014, Card et al. 2018) by addressing two major questions: 1) how concentrated are local labour markets? and 2) what is the impact of concentration on wages? We believe we are the first to address these questions by exploiting rich matched employer-employee data. Our data covers the full population of workers (and occupations) in a country, Portugal, including not only manufacturing but also services as well as both incumbent workers and new hires. Furthermore, in contrast to other papers in this literature, our data also includes information on the occupation, region and wages of each worker. All these variables are particularly detailed and comparable not only across firms but also over time, during the 23-year period that we cover. These two data dimensions - coverage of both all occupations and of the entire labour market - are critical for a comprehensive analysis of employer market power. The majority of workers in developed countries have been for many decades employed in the services sector (which is however not covered in a number of data sets used in this research) and concentration levels may be much higher in manufacturing than services. Moreover, a less than comprehensive knowledge of the spread of comparable occupations across firms as well as an analysis based exclusively on employment flows or stocks may also lead to overestimated measures of concentration. A key reference in this literature is Azar et al. (2017) (see also Azar et al. (2018)): using data on the top occupations from a leading U.S. employment website, they find that the average labour market is highly concentrated. Moreover, using a subset of their data for which posted wage information is available, they find that concentration is associated with 2 large declines in earnings: cuts of 17% from moving from the 25th to the 75th percentile of the Herfindhal distribution.1 Our results are consistent with the previous literature in that we also find that employer market power is a significant phenomenon. However, our results indicate that employer market power effects are smaller than in earlier evidence that could not draw on data with the same level of detail. This finding of smaller effects applies both in terms of the percentage of the workforce employed in high-concentration local labour markets and in terms of the size of the wage effects of concentration. Specifically, we find that less than 9% of workers are exposed to concentration levels thought to raise market power concerns (DoJ/FTC 2010). However, we also show that these measures can more than double with different methodological choices, in particular when considering more detailed occupation codes and more disaggregated geographical areas. Our wage analysis highlights the relevance of controlling for worker (and firm) composi- tion. When not doing so, we find a positive association between employer concentration and wages. This may reflect the fact that concentrated labour markets tend to be characterised by large firms, which will tend to be more productive and or enjoy higher product market power. These two dimensions will in turn create scope both for stronger selection in hirings (em- ploying workers of higher productivity) as well as rent sharing, through bilateral bargaining, particularly when labour market policies are more supportive of collective bargaining. We address these factors by again exploiting the richness of our data and estimating the relationship between wages and concentration using both worker and firm fixed effects. We follow closely Azar et al. (2017) and employ instrumental variables based on the number of firms in the same occupations but other local labour markets. In these models, we find a negative effect of local labour market concentration on wages, with elasticities of -1.5%. These results indicate that workers that would move from low- to high-concentration local labour markets (percentiles 25th and 75th, respectively) would experience a drop in wages of approximately 2%, a figure significantly smaller than in Azar et al. (2017). This smaller effect may reflect the role of different labour market institutions, in particular sectoral collective bargaining, which is much stronger in Portugal than the U.S, and can erode the negative 1See also Benmelech et al. (2018) and Rinz (2018) for two recent contributions focusing on the case of the U.S and Duan & Martins (2019) who also examine the role of employer labour market power in shaping rent sharing in China. The last two papers examine the manufacturing sector only. The three papers examine data that does not provide information on occupations per firm. 3 effect of employer market power. The structure of the remaining of the paper is as follows: Section 2 presents the data used. Sections 3 and 4 describe our main measures of employer concentration and a comprehensive sensitivity analysis that we also conduct. The wage effects results are presented in Section 5. Finally, Section 6 concludes. 2 Data Our empirical study is based on the `Quadros de Pessoal' (Personnel Records) data set, a comprehensive matched employer-employee panel. This data set provides detailed annual information on all firms based in Portugal that employ at least one worker, including individual information on each of their employees and time-invariant firm and worker identifiers. The data set follows from an annual mandatory survey collected by the Ministry of Employment for the purposes of monitoring and enforcing compliance with employment law. Worker information concerns the month of October of each year and features a number of variables, on an individual basis, the most important for us being the monthly wage (base and total) and the occupation. The latter is defined using the national occupations classification, which features about 1,400 different entries, defined at a six-digit level (the national occupation classification, unchanged between 1995 and 2009). Other worker-level variables that we also consider include the year and month of birth, the year and month of hiring by the firm, the applicable collective agreement, and the job title code of the worker under the collective agreement. At the firm-level, we use the geographical location of the firm (at the 'concelho' level - over 450 different locations - and the more aggregated 'distrito' level - 30 different locations, including the islands of Azores and Madeira).2 Our benchmark definition of local labour markets is based on the occupation definition composed of 1,400 entry and the region definition composed of 30 entries ('distritos'). These regions have an average size of about 3,000 square kilometres and 340,000 inhabitants, including multiple cities and villages in most cases. 18 regions are located in the continental part of Portugal; the remaining 12 correspond to the islands of Azores and Madeira. With the exception of some areas in the least populated and least densely populated regions, typically 2We use all worker observations in the data set except the small number with missing information on the occupation, date of hiring, and region variables. Other worker-level variables available are gender, schooling, hours of work, and type of employment contract. At the firm-level, the data also provides information on industry (five-digit variable), total sales, legal type of firm, capital equity, and type of ownership. 4 located in the East and South of the country, all urban or industrial areas can be reached from virtually all residential areas in each district in less than one hour by car and in many cases in less than one hour by public transport as well.

Load more